Where Does Speaker Structure Belong in AI Podcast Scripts?

Every multi-speaker AI pipeline runs on a contract with no enforcement layer — and when it breaks, the failure gets blamed on the model. This episode digs into a documented bug report from an open-source podcast generator: six segments, three speakers, forty-two dialogue lines, and a model that returned the prompt's own JSON skeleton verbatim. Then it turns to the obvious fix — strict structured output — and what two research papers actually found about the cost of asking a model to respond in JSON. The answer isn't where most people assume: the damage happens at the prompt, not the decoder. Along the way: why high-temperature sampling doesn't restore diversity under a template, why capacity matters more than format for strong models, and why "thinking first" helps math but does nothing for prose. Key sources: • open-notebook GitHub issue #1238, Podcast generation — https://github.com/lfnovo/open-notebook/issues/1238 • Together AI docs, Open NotebookLM: PDF to Podcast (Pydantic `LineItem`/`Script` schema, JSON mode). — https://docs.together.ai/docs/open-notebooklm-pdf-to-podcast • The Format Tax — https://arxiv.org/html/2604.03616 Episode 6056 — myweirdprompts.com/n/6056 (permanent short link: myweirdprompts.com/786203) Listen on Android: MWP FM Player — myweirdprompts.com/apk